# Calibrated harness: live mc5 fills vs 5-Min Cap replay (testmgr, 10 Oct, #371)

Data: fw/mc5live.txt (19 live BUYs since 8 Oct, 18 closed) against the grid replay's live-rule run (LIVE 0.5x/4h/250k-1M/5m, tick-walk on real pool depth, 1% each way + gas). 15 coins matched by token.

## Result: the replay is already close to live
| item | live | replay | gap |
|---|---|---|---|
| mean x, 15 matched trades | 1.235 | 1.140 | replay 8% low, all from the one 12.8x runner (replay 11.3x) |
| stop exits, 14 matched, mean x | 0.409 | 0.414 | -0.005 (median +0.007): no bias |
| stop fill vs the 0.5 trigger | mean 0.41, median 0.48 | 0.41 / 0.46 | both below 0.5 on average |
| gap-through stops (x < 0.3) | 3 of 17 (0.01, 0.11, 0.29) | 2 of 14 (0.08, 0.11) | replay has most of them |

## Fitted live parameters (18 closed)
- Entry: fill delay mean 3.7 s (median 3.3, max 12 s). Fill/sim mean 1.029, median 1.00 (range 0.94-1.17): entry slip is small, about +3% worst case; the 5% slippage cap is rarely used.
- Stop fill: median 0.48x of entry, mean 0.41x because of rare gap-through (see above). A flat 0.5 stop (as in simple sims) overstates stop fills by about 18%; the tick-walk replay does not.
- Sell gap (pool price vs fill) median +0.002, mean +0.011; stop slip median -0.008, mean -0.075 (driven by the 0.01x rug where the pool fell through).
- Exit error on the 12.8x runner (time exit 4 h): live 12.8x vs replay 11.3x, i.e. replay is conservative at the top by 12%.

## What to use in other tests
- Entry slip 3% mean on top of pool fee; delay 3-4 s.
- Stop fill: use tick-walk on real depth, never a flat trigger price. Where depth is not available use 0.48 with 15% chance of 0.1.
- Fast-rug pools (stop 0.01-0.11): 3 of 18 trades, roughly 15%; they set the loss side.

## Caveats
n is 18 closed and one runner is 100% of the profit over 18 trades; this fits slip and stop fills, not the edge. Only 15 of 18 had a replay row (3 newest not yet replayed). Quote-fail misses (skip_quote on fml2) are a separate loss, see tm-miss-skipquote-10oct.md.